Neural Network Gives Insight Into Endoscopic Images for Ulcerative Colitis

March 07, 2020
Evaluating endoscopic images from patients with ulcerative colitis (UC) remains a challenging endeavor.

A team, led by Kento Takenaka, PhD, Department of Gastroenterology and Hepatology at Tokyo Medical and Dental University, developed a new deep neural network system for consistent, objective, and real-time analysis of endoscopic images from patients with ulcerative colitis.

The investigators constructed the deep neural network using 40,758 images of colonoscopies and 6885 biopsy results from 2012 patients with ulcerative colitis who underwent colonoscopies from January 2014 to March 2018 at a single center in Japan.

The team validated the accuracy of the deep neural network algorithm in a prospective study involving 875 ulcerative colitis patients who underwent a colonoscopy from April 2018 through April 2019. This included 4187 endoscopic images and 4104 biopsy specimens.

The investigators defined endoscopic remission as an UC endoscopic index of severity (UCEIS) score of 0, as well as a histologic remission defined as a Geboes score of 3 points of less.

The new deep neural network identified with a 90.1% accuracy patients with endoscopic remission (95% CI, 89.2–90.9%), as well as a kappa coefficient of 0.798 (95% CI, 0.780–0.814) by using the findings reported by endoscopists as the reference standard. The intraclass correlation coefficient between the neural network and the endoscopists for UCEIS scoring was 0.917 (95% CI, 0.911–0.921).

The system identified patients in histologic remission with a 92.9% accuracy (95% CI, 92.1–93.7%), while the kappa coefficient between the DNUC and the biopsy result was 0.859 (95% CI, 0.841–0.875).

“We developed a deep neural network for evaluation of endoscopic images from patients with UC that identified those in endoscopic remission with 90.1% accuracy and histologic remission with 92.9% accuracy,” the authors wrote. “The DNUC can therefore identify patients in remission without the need for mucosal biopsy collection and analysis.” 

Earlier this year, investigators presented a new prognostic model could help identify the one-year risk of Crohn’s disease-related intestinal surgery.

A team, led by Jia-Yin Yao, Department of Gastroenterology, Guangdong Provincial Key Laboratory of Colorectal and Pelvic Floor Diseases, the Sixth Affiliated Hospital of Sun Yat-Sen University, collected data from Crohn’s disease patients diagnosed between January 2012 and December 2016 and developed a prognostic model that predicts the one-year surgery risk for patients.

In the retrospective study, the investigators randomly stratified all the data into a training set and a testing set at a ratio of 8:2.

A number of variables were identified as independent significant factors associated with early intestinal surgery, including disease behavior (B2: OR, 6.693; P <.001; B3: OR, 14.405; P <.001), smoking (OR, 4.135; P <.001), body mass index (OR, .873, P <.001) and C-reactive protein (OR, 1.022; P = .001) at diagnosis, previous perianal (OR, 9.483; P <.001) or intestinal surgery (OR, 8.887; P <.001), maximum bowel wall thickness (OR, 1.965; P <.001), use of biologics (OR, .264; P <.001), and exclusive enteral nutrition (OR, .089; P <.001).

To further validate this, the investigators established the prognostic model, where the receiver operating characteristic curves and calculated areas under the curves (94.7%) confirmed an ideal predictive ability of the model with a sensitivity of 75.92% and specificity of 95.81%.

“This prognostic model can effectively predict 1-year risk of [Crohn’s disease-related] intestinal surgery, which will assist in screening progressive [Crohn’s disease] patients and tailoring therapeutic management,” the authors wrote.

The study, “Development and Validation of a Deep Neural Network for Accurate Evaluation of Endoscopic Images From Patients with Ulcerative Colitis,” was published online in Gastroenterology.
 
Copyright® ConsultantLive 2019 Intellisphere, LLC. All Rights Reserved.